CN111060652A - Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network - Google Patents
Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network Download PDFInfo
- Publication number
- CN111060652A CN111060652A CN201911145374.1A CN201911145374A CN111060652A CN 111060652 A CN111060652 A CN 111060652A CN 201911145374 A CN201911145374 A CN 201911145374A CN 111060652 A CN111060652 A CN 111060652A
- Authority
- CN
- China
- Prior art keywords
- concentration
- gas
- sequence
- oil
- dissolved gas
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000007787 long-term memory Effects 0.000 title claims abstract description 10
- 230000006403 short-term memory Effects 0.000 title claims abstract description 10
- 239000007789 gas Substances 0.000 claims abstract description 162
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000005065 mining Methods 0.000 claims abstract description 9
- 239000003921 oil Substances 0.000 claims description 56
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims description 20
- 229910002092 carbon dioxide Inorganic materials 0.000 claims description 20
- 230000000875 corresponding effect Effects 0.000 claims description 12
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims description 11
- 229910002091 carbon monoxide Inorganic materials 0.000 claims description 11
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 claims description 10
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 claims description 10
- 239000005977 Ethylene Substances 0.000 claims description 10
- 239000001569 carbon dioxide Substances 0.000 claims description 10
- 229930195733 hydrocarbon Natural products 0.000 claims description 10
- 150000002430 hydrocarbons Chemical class 0.000 claims description 10
- 229910052739 hydrogen Inorganic materials 0.000 claims description 10
- 239000001257 hydrogen Substances 0.000 claims description 10
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000003786 synthesis reaction Methods 0.000 claims description 6
- 230000002596 correlated effect Effects 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000015654 memory Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0036—General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0036—General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
- G01N33/004—CO or CO2
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0036—General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
- G01N33/005—H2
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0062—General constructional details of gas analysers, e.g. portable test equipment concerning the measuring method or the display, e.g. intermittent measurement or digital display
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Immunology (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Medicinal Chemistry (AREA)
- Food Science & Technology (AREA)
- Combustion & Propulsion (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting the concentration of dissolved gas in transformer oil based on a long-term and short-term memory network, which comprises the following steps: the method comprises the following steps: collecting historical measured data of the concentration of the dissolved gas in the transformer oil; step two: mining the association relation between dissolved gases in oil by using an association rule method to obtain an association rule between gas concentration sequences; step three: utilizing the concentration sequence of the dissolved gas in the oil in the wavelet decomposition processing step I to obtain a low-frequency sequence component and a high-frequency sequence component of the concentration sequence of the dissolved gas in the oil; step four: respectively predicting the dissolved gas sequence components in the oil in the third step by using the LSTM, and then recombining the predicted dissolved gas sequence components; using root mean square error eRMSEAnd the mean absolute error eMAETwo indexes calculate and predict errorAnd (4) poor. The method can better track the concentration change trend of the dissolved gas in the oil, and has higher prediction precision.
Description
Technical Field
The invention belongs to the field of transformer gas concentration prediction, and particularly relates to a method for predicting the concentration of dissolved gas in transformer oil based on a long-term and short-term memory network.
Background
The power transformer is an important device for voltage lifting and power distribution in a power system, and the normal operation of the power transformer is related to the safety and stability of the whole power grid. During the operation and use of the transformer, a small amount of gas is generated and dissolved in the insulating oil due to aging, electric and thermal faults and the like, and the volume fractions of various components and the proportional relation among different components of the gas in the oil are closely related to the operation condition of the transformer. Analysis of dissolved gas in transformer oil is an important method for diagnosing early latent faults of transformers, which is most widely applied at present. Through analyzing the historical monitoring sequence, the development trend of the concentration of the dissolved gas in the oil is accurately predicted, the running condition of the transformer can be mastered in advance, and a basis can be provided for the state evaluation and the state maintenance of the transformer.
The method mainly comprises a statistical prediction method and a machine learning method. The statistical prediction method mainly comprises a time sequence model and a gray model, the prediction accuracy of the statistical prediction method is limited by the distribution rule of the sequence, and the applicable scene has great limitation. The traditional machine learning method comprises a support vector machine, an artificial neural network and the like, and a prediction model capable of reflecting the development trend of a time series is obtained by analyzing and training a large amount of historical data. However, the traditional machine learning method only predicts through historical data, ignores the correlation between the concentrations of the dissolved gases in the oil, and the prediction accuracy is to be improved.
Disclosure of Invention
In order to solve the problems, the invention provides a method for predicting the concentration of dissolved gas in transformer oil based on a long-term and short-term memory network.
The technical scheme for realizing the purpose of the invention is as follows:
the method for predicting the concentration of dissolved gas in transformer oil based on the long-term and short-term memory network comprises the following steps:
the method comprises the following steps: collecting historical measured data of dissolved gas concentration in transformer oil, including ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Total hydrocarbons, establishing a gas concentration sequence of the 7 gases in order of date of data acquisition;
step two: mining the association relation between dissolved gases in oil by using an association rule method to obtain an association rule between gas concentration sequences;
step three: utilizing the concentration sequence of the dissolved gas in the oil in the wavelet decomposition processing step I to obtain a low-frequency sequence component and a high-frequency sequence component of the concentration sequence of the dissolved gas in the oil;
step four: respectively predicting the dissolved gas sequence components in the oil in the third step by using the LSTM, and then recombining the predicted dissolved gas sequence components, including
Training the transformer running state prediction model by adopting a time-back propagation algorithm, mining the correlation relationship between dissolved gases in oil according to a correlation rule, taking a gas concentration sequence correlated with the concentration of the gas to be predicted and a subsequence decomposed by the gas concentration sequence to be predicted as input variables, constructing n LSTM prediction models, respectively predicting the next time low-frequency sequence component and high-frequency sequence component of each layer of sequence, and then performing wavelet reconstruction synthesis on the predicted values of the low-frequency sequence component and the high-frequency sequence component at each moment, wherein the wavelet reconstruction synthesis formula is
n is the number of wavelet decomposition layers;
step five: using root mean square error eRMSEAnd the mean absolute error eMAEThe two indexes calculate the prediction error according to the formula
Wherein, yi、
The real value and the predicted value of the concentration of the dissolved gas in the oil are respectively shown, n represents the number of the test data, and i represents the serial number of the prediction point.
Further, in the second step, the association relation between the dissolved gases in the oil is mined by using an association rule method, and a specific method for obtaining the association rule between the gas concentration sequences is as follows:
firstly, carrying out single normalization processing on the 7 gas concentration sequences in the step one to obtain all normalization values of the gas concentration sequences, wherein all the normalization values are between 0 and 1, and the calculation formula is as follows:
in the formula, max is ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2A maximum concentration of one gas in the total hydrocarbons, min is the minimum concentration of the corresponding gas, xiIs the gas concentration sequence value of the corresponding gas, and j is the number of samples of the collected gas concentration;
discretizing the normalized data by adopting a partitioning method based on k clustering, symbolizing a clustering result, and expressing the clustering result by using 'A', 'B', 'C', 'D' and … …, wherein a clustering formula is as follows:
in the formula, xmIs ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Value of a series of gas concentrations in total hydrocarbons, muiThe mean value of the ith cluster of the corresponding gas concentration is the Euclidean distance, k is the cluster category number, and n is the sample number of the corresponding gas concentration;
and finding out a frequent item set with the support degree of the gas concentration item set greater than the minimum support degree by using an Apriori algorithm, and deleting a rule with the confidence degree of the frequent item set less than a threshold value to obtain the association relation among the gas concentration sequences.
Further, an Apriori algorithm is used for finding out a frequent item set with the gas concentration item set with the support degree larger than the minimum support degree, a rule that the confidence degree in the frequent item set is smaller than a threshold value is deleted, and the association relation between the gas concentration sequences is obtained, wherein the process is as follows:
the concentration data of dissolved gas in seven oils represented by the normalized symbols are represented as D, D ═ t1,t2,...,tnWhere tk＝{i1,i2,...,in}，tk(k ═ 1, 2.., n) is referred to as a transaction, im(m ═ 1,2,.. p) is referred to as a term. Scanning all affairs in the dissolved gas concentration data set in oil, respectively calculating the times of clustering centers 'A', 'B', 'C', 'D' and … …, finding out a frequent item set, wherein the correlation degree calculation formula among gas concentration sequences is as follows:
x, Y is called front piece and back piece of the association rule respectively, count (X ∩ Y) is the number of X and Y contained in database D at the same time, only when the support degree of the rule is greater than the set minimum support degree, the item set is called a frequent item set, the minimum support degree is generally a set value, and the calculation formula of the association rule credibility among gas concentration sequences is
If there are n rules X for the gas concentration sequence X and the gas concentration sequence Yi→YiIf the association rule satisfies the minimum confidence, using the formula:
and (4) calculating the correlation degree and confidence degree among the gas concentration sequences, and finding out the sequences with strong correlation.
Further, in the third step, the specific method for obtaining the low-frequency sequence component and the high-frequency sequence component of the concentration sequence of the dissolved gas in the oil by using the concentration sequence of the dissolved gas in the oil in the wavelet decomposition processing step one is as follows:
orthogonal projection of signal x (t) into space V using the fast algorithm Mallet algorithm of Daubechies waveletsjAnd WjRespectively obtaining discrete approximation signals c under the resolution jj(t) and a discrete detail signal dj(t) increasing j from zero step by step to realize the step-by-step decomposition of the signal, decomposing the low-frequency component obtained by the last decomposition into a low-frequency part and a high-frequency part by the result of each step of decomposition, and not considering the high-frequency signal to obtain subsequences which are respectively the low-frequency component an(t), high frequency component dj(t); the signal x (t) after multi-resolution decomposition can be represented as
And n is the number of wavelet decomposition layers.
The invention has the beneficial effects that:
the invention provides a method for predicting the concentration of dissolved gas in transformer oil based on wavelet decomposition (WT) and long-short term memory (LSTM) aiming at the defects of the existing prediction method. According to the method, firstly, state parameters strongly associated with the gas to be predicted are mined through association rules, then the gas sequence to be predicted is decomposed into subsequences with different frequencies by utilizing wavelet transform, low-frequency trend components and high-frequency fluctuation components in the sequence are extracted, then LSTM is used for predicting on different components by combining with the association rules among gas sequences dissolved in oil, and the prediction result of the gas sequence to be predicted is obtained through reconstruction. The method is used for predicting the dissolved gas in 220kV main transformer oil of a certain transformer substation, and the result shows that the prediction precision of the method is higher compared with a prediction method without considering relevance and a traditional prediction method.
Drawings
FIG. 1 is a graph showing the results of carbon monoxide distribution.
Fig. 2 is a diagram of the wavelet decomposition result of the CO sequence.
FIG. 3 is a graph of the CO concentration prediction results based on WT-LSTM.
FIG. 4 is a graph of WT-LSTM CO concentration prediction results taking into account the gas concentration dependence.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments that can be derived by one of ordinary skill in the art from the embodiments given herein are intended to be within the scope of the present invention.
According to an embodiment of the invention, a method for predicting the concentration of dissolved gas in transformer oil based on a long-short term memory network is provided, and the method for predicting the concentration of dissolved gas in transformer oil based on wavelet decomposition and the long-short term memory network comprises the following steps:
the method comprises the following steps: collecting historical measured data of dissolved gas concentration in transformer oil, including ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Total hydrocarbons, establishing a gas concentration sequence of the 7 gases in order of date of data acquisition;
step two: the method for mining the association relation between the dissolved gases in the oil by using an association rule method to obtain the association rule between the gas concentration sequences comprises the following steps:
firstly, carrying out single normalization processing on the 7 gas concentration sequences in the step one to obtain all normalization values of the gas concentration sequences, wherein all the normalization values are between 0 and 1, and the calculation formula is as follows:
in the formula, max is ethane C2H6Hydrogen gasH2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2A maximum concentration of one gas in the total hydrocarbons, min is the minimum concentration of the corresponding gas, xiIs the gas concentration sequence value of the corresponding gas, and j is the number of samples of the collected gas concentration;
discretizing the normalized data by adopting a partitioning method based on k clustering, symbolizing a clustering result, and expressing the clustering result by using 'A', 'B', 'C', 'D' and … …, wherein a clustering formula is as follows:
in the formula, xmIs ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Value of a series of gas concentrations in total hydrocarbons, muiThe mean value of the ith cluster of the corresponding gas concentration is the Euclidean distance, k is the cluster category number, and n is the sample number of the corresponding gas concentration;
finding out a frequent item set with the support degree of the gas concentration item set greater than the minimum support degree by using an Apriori algorithm, deleting a rule that the confidence degree of the frequent item set is less than a threshold value, and obtaining an association relation among the gas concentration sequences, wherein the association relation comprises the following steps:
the concentration data of dissolved gas in seven oils represented by the normalized symbols are represented as D, D ═ t1,t2,…,tnWhere tk＝{i1,i2,…,in}，tk(k ═ 1,2, …, n) is referred to as a transaction, im(m ═ 1,2, …, p) is referred to as a term. Scanning all affairs in the dissolved gas concentration data set in oil, respectively calculating the times of clustering centers 'A', 'B', 'C', 'D' and … …, finding out a frequent item set, wherein the correlation degree calculation formula among gas concentration sequences is as follows:
wherein X, Y are respectively calledThe front part and the back part of the association rule, count (X ∩ Y) is the number of X and Y contained in the database D at the same time, only when the support degree of the rule is greater than the set minimum support degree, the item set is called a frequent item set, the minimum support degree is generally a set value, and the calculation formula of the credibility of the association rule among the gas concentration sequences is that
If there are n rules X for the gas concentration sequence X and the gas concentration sequence Yi→YiIf the association rule satisfies the minimum confidence, using the formula:
and (4) calculating the correlation degree and confidence degree among the gas concentration sequences, and finding out the sequences with strong correlation.
Step three: utilizing the concentration sequence of the dissolved gas in the oil in the wavelet decomposition processing step I to obtain a low-frequency sequence component and a high-frequency sequence component of the concentration sequence of the dissolved gas in the oil;
orthogonal projection of signals x (t) into space V using the fast algorithm Mallet algorithm of Daubechies (dbN) waveletsjAnd WjRespectively obtaining discrete approximation signals c under the resolution jj(t) and a discrete detail signal dj(t) increasing j from zero step by step to realize the step-by-step decomposition of the signal, decomposing the low-frequency component obtained by the last decomposition into a low-frequency part and a high-frequency part by the result of each step of decomposition, and not considering the high-frequency signal to obtain subsequences which are respectively the low-frequency component an(t), high frequency component dj(t); the signal x (t) after multi-resolution decomposition can be represented as
And n is the number of wavelet decomposition layers.
Step four: respectively predicting the dissolved gas sequence components in the oil in the third step by using the LSTM, and then recombining the predicted dissolved gas sequence components, including
By reversing in timeTraining the transformer running state prediction model by a propagation algorithm, mining the correlation relationship between dissolved gases in oil according to the correlation rule, taking a gas concentration sequence correlated with the concentration of the gas to be predicted and a subsequence decomposed by the gas concentration sequence to be predicted as input variables, constructing n LSTM prediction models, respectively predicting the low-frequency sequence component and the high-frequency sequence component of each layer of sequence at the next time, and then performing wavelet reconstruction synthesis on the predicted values of the low-frequency sequence component and the high-frequency sequence component at each moment, wherein the wavelet reconstruction synthesis formula is
n is the number of wavelet decomposition layers;
step five: using root mean square error eRMSEAnd the mean absolute error eMAEThe two indexes calculate the prediction error according to the formula
Wherein, yi、
The real value and the predicted value of the concentration of the dissolved gas in the oil are respectively shown, n represents the number of the test data, and i represents the serial number of the prediction point.
The present invention will be further described with reference to specific examples.
The invention provides a method for realizing the concentration prediction of dissolved gas in transformer oil of a wavelet decomposition and long-short term memory network, which comprises the following concrete realization processes:
1. and collecting the concentration data of the dissolved gas in the transformer oil. In the embodiment of the invention, measured data of a certain 220kV transformer is collected, sample data is transformer oil chromatogram online monitoring data from 11/19/2017 to 7/4/2019, and the length of a data set is 547.
2. And excavating the association relation between the dissolved gases in the oil by utilizing the association rule.
Para ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Normalizing the total hydrocarbons;
taking CO data as an example, the result of normalizing the data is shown in fig. 1:
discretizing by utilizing k clustering, wherein k is 4, and discretizing results are as follows:
TABLE 1 discretization results
Mining the correlation among the gas concentrations by using an Apriori algorithm, setting the support degree and the confidence degree to be 0.5, and obtaining the following mining results as shown in the following table 2:
table 2 association rules mined
3. And processing the gas sequence to be predicted by utilizing wavelet decomposition.
Taking the CO concentration as an example, the wavelet decomposition is carried out;
according to the method, a tightly-supported biorthogonal db6 wavelet is selected to carry out 3-layer wavelet decomposition on a CO signal, an original sequence and a subsequence obtained after decomposition are shown in figure 2, and therefore, the CO sequence is decomposed into 4 subsequences;
4. gas concentration prediction was performed using LSTM.
The method adopts python to realize LSTM gas concentration prediction programming;
predicted realizations of CO concentrations without taking into account correlations between gases.
Respectively taking each subsequence component of CO as an input variable, constructing 4 LSTM neural network prediction models, reconstructing the prediction result of each subsequence to obtain the final prediction value of CO, wherein the result is shown in figure 3
A predictive implementation of gas dependence is considered.
2, in the correlation between the dissolved gases in the oil, the CO is known by using the correlation rule2、H2、C2H4The concentration and the CO concentration have strong correlation, so that in the section, historical data of the three sequences and subsequences obtained after CO decomposition are respectively used as input variables, then 4 multi-input LSTM neural network prediction models considering correlation rules are constructed, finally, prediction results of the models are reconstructed to obtain final predicted values of the CO, and the results are shown in FIG. 4;
in order to analyze the effectiveness of the prediction model, the prediction results obtained by the Elman neural network prediction, the dynamic neural network NAR prediction and the LSTM prediction are compared and analyzed with the prediction results obtained by the method, and the prediction results used for the comparison method are shown in the following table 3;
TABLE 3 comparison of predicted Performance
As can be seen from the above table, the WT-LSTM model has e compared to the Elman, NAR and LSTM modelsRMSEIndex decreases by 23.7%, 24% and 16.3%, respectively, eMAEThe reduction is respectively 26.2%, 26.6% and 17.1%, and therefore, the smoothing processing of the sequence by utilizing the wavelet transform is beneficial to improving the prediction precision. While the AR-WT-LSTM model, which takes into account the inter-sequence association rules, compares to the simple WT-LSTM model, eRMSEIndex is reduced by 6%, eMAEThe method reduces 12.4%, further improves the prediction accuracy, and verifies the effectiveness of the introduced association rule in the transformer operation parameter prediction.
Claims (4)
1. The method for predicting the concentration of dissolved gas in transformer oil based on the long-term and short-term memory network is characterized by comprising the following steps of:
the method comprises the following steps: collecting historical measured data of dissolved gas concentration in transformer oil, including ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Total hydrocarbons, establishing a gas concentration sequence of the 7 gases in order of date of data acquisition;
step two: mining the association relation between dissolved gases in oil by using an association rule method to obtain an association rule between gas concentration sequences;
step three: utilizing the concentration sequence of the dissolved gas in the oil in the wavelet decomposition processing step I to obtain a low-frequency sequence component and a high-frequency sequence component of the concentration sequence of the dissolved gas in the oil;
step four: respectively predicting the dissolved gas sequence components in the oil in the third step by using the LSTM, and then recombining the predicted dissolved gas sequence components, including
Training the transformer running state prediction model by adopting a time-back propagation algorithm, mining the correlation relationship between dissolved gases in oil according to a correlation rule, taking a gas concentration sequence correlated with the concentration of the gas to be predicted and a subsequence decomposed by the gas concentration sequence to be predicted as input variables, constructing n LSTM prediction models, respectively predicting the next time low-frequency sequence component and high-frequency sequence component of each layer of sequence, and then performing wavelet reconstruction synthesis on the predicted values of the low-frequency sequence component and the high-frequency sequence component at each moment, wherein the wavelet reconstruction synthesis formula is
n is the number of wavelet decomposition layers;
step five: using root mean square error eRMSEAnd the mean absolute error eMAEThe two indexes calculate the prediction error according to the formula
Wherein, yi、
The real value and the predicted value of the concentration of the dissolved gas in the oil are respectively shown, n represents the number of the test data, and i represents the serial number of the prediction point.
2. The method for predicting the concentration of the dissolved gas in the transformer oil based on the long and short term memory network as claimed in claim 1, wherein in the second step, the association relation between the dissolved gas in the oil is mined by using an association rule method, and the specific method for obtaining the association rule between the gas concentration sequences is as follows:
firstly, carrying out single normalization processing on the 7 gas concentration sequences in the step one to obtain all normalization values of the gas concentration sequences, wherein all the normalization values are between 0 and 1, and the calculation formula is as follows:
i is more than or equal to 0 and less than or equal to j, in the formula, max is ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2A maximum concentration of one gas in the total hydrocarbons, min is the minimum concentration of the corresponding gas, xiIs the gas concentration sequence value of the corresponding gas, and j is the number of samples of the collected gas concentration;
discretizing the normalized data by adopting a partitioning method based on k clustering, symbolizing a clustering result, and expressing the clustering result by using 'A', 'B', 'C', 'D' and … …, wherein a clustering formula is as follows:
in the formula, xmIs ethane C2H6Hydrogen gas H2Methane CH4Ethylene C2H4Carbon monoxide CO and carbon dioxide CO2Value of a series of gas concentrations in total hydrocarbons, muiThe mean value of the ith cluster of the corresponding gas concentration is the Euclidean distance, k is the cluster category number, and n is the sample number of the corresponding gas concentration;
and finding out a frequent item set with the support degree of the gas concentration item set greater than the minimum support degree by using an Apriori algorithm, and deleting a rule with the confidence degree of the frequent item set less than a threshold value to obtain the association relation among the gas concentration sequences.
3. The method for predicting the concentration of dissolved gas in transformer oil based on the long-term and short-term memory network as claimed in claim 2, wherein an Apriori algorithm is used to find out a frequent item set with a gas concentration item set supporting degree greater than a minimum supporting degree, and a rule that a central reliability of the frequent item set is less than a threshold value is deleted to obtain the correlation between gas concentration sequences, and the process is as follows:
the concentration data of dissolved gas in seven oils represented by the normalized symbols are represented as D, D ═ t1,t2,...,tnWhere tk＝{i1,i2,...,in}，tk(k ═ 1, 2.., n) is referred to as a transaction, im(m ═ 1,2,.. p) is referred to as a term. Scanning all affairs in the dissolved gas concentration data set in oil, respectively calculating the times of clustering centers 'A', 'B', 'C', 'D' and … …, finding out a frequent item set, wherein the correlation degree calculation formula among gas concentration sequences is as follows:
x, Y is called front piece and back piece of the association rule respectively, count (X ∩ Y) is the number of X and Y contained in database D at the same time, only when the support degree of the rule is greater than the set minimum support degree, the item set is called a frequent item set, the minimum support degree is generally a set value, and the calculation formula of the association rule credibility among gas concentration sequences is
If there are n rules X for the gas concentration sequence X and the gas concentration sequence Yi→YiIf the association rule satisfies the minimum confidence, using the formula:
and (4) calculating the correlation degree and confidence degree among the gas concentration sequences, and finding out the sequences with strong correlation.
4. The method for predicting the concentration of the dissolved gas in the transformer oil based on the long and short term memory network as claimed in claim 2 or 3, wherein in the third step, the sequence of the concentration of the dissolved gas in the oil in the first wavelet decomposition processing step is used, and the specific method for obtaining the low frequency sequence component and the high frequency sequence component of the sequence of the concentration of the dissolved gas in the oil is as follows:
orthogonal projection of signal x (t) into space V using the fast algorithm Mallet algorithm of Daubechies waveletsjAnd WjRespectively obtaining discrete approximation signals c under the resolution jj(t) and a discrete detail signal dj(t) increasing j from zero step by step to realize the step-by-step decomposition of the signal, decomposing the low-frequency component obtained by the last decomposition into a low-frequency part and a high-frequency part by the result of each step of decomposition, and not considering the high-frequency signal to obtain subsequences which are respectively the low-frequency component an(t), high frequency component dj(t); the signal x (t) after multi-resolution decomposition can be represented as
And n is the number of wavelet decomposition layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911145374.1A CN111060652A (en) | 2019-11-21 | 2019-11-21 | Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911145374.1A CN111060652A (en) | 2019-11-21 | 2019-11-21 | Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111060652A true CN111060652A (en) | 2020-04-24 |
Family
ID=70298531
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911145374.1A Pending CN111060652A (en) | 2019-11-21 | 2019-11-21 | Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111060652A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112734028A (en) * | 2020-12-28 | 2021-04-30 | 三峡大学 | Modeling method for prediction model of concentration of dissolved gas in transformer oil |
| CN112749509A (en) * | 2020-12-30 | 2021-05-04 | 西华大学 | Intelligent substation fault diagnosis method based on LSTM neural network |
| CN113504423A (en) * | 2021-07-13 | 2021-10-15 | 许昌许继软件技术有限公司 | Primary equipment online monitoring data trend prediction method and device |
| CN113705086A (en) * | 2021-08-05 | 2021-11-26 | 陶帝文 | Ultra-short-term wind power prediction method based on Elman error correction |
| CN117667495A (en) * | 2023-12-29 | 2024-03-08 | 湖北华中电力科技开发有限责任公司 | Application system fault prediction method based on association rule and deep learning integrated model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106384170A (en) * | 2016-09-24 | 2017-02-08 | 华北电力大学（保定） | Wavelet decomposition and reconstruction-based time sequence wind speed prediction method |
| CN106447086A (en) * | 2016-09-07 | 2017-02-22 | 中国农业大学 | Wind electricity power combined prediction method based on wind farm data pre-processing |
| CN110018670A (en) * | 2019-03-28 | 2019-07-16 | 浙江大学 | A kind of industrial process unusual service condition prediction technique excavated based on dynamic association rules |
-
2019
- 2019-11-21 CN CN201911145374.1A patent/CN111060652A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447086A (en) * | 2016-09-07 | 2017-02-22 | 中国农业大学 | Wind electricity power combined prediction method based on wind farm data pre-processing |
| CN106447086B (en) * | 2016-09-07 | 2019-09-24 | 中国农业大学 | One kind being based on the pretreated wind power combination forecasting method of wind farm data |
| CN106384170A (en) * | 2016-09-24 | 2017-02-08 | 华北电力大学（保定） | Wavelet decomposition and reconstruction-based time sequence wind speed prediction method |
| CN110018670A (en) * | 2019-03-28 | 2019-07-16 | 浙江大学 | A kind of industrial process unusual service condition prediction technique excavated based on dynamic association rules |
Non-Patent Citations (3)
| Title |
|---|
| 代杰杰 等: "考虑复杂关联关系深度挖掘的变压器状态参量预测方法", 《中国电机工程学报》 * |
| 刘云鹏 等: "基于经验模态分解和长短期记忆神经网络的变压器油中溶解气体浓度预测方法", 《中国电机工程学报》 * |
| 林峻 等: "考虑时间序列关联的变压器在线监测数据清洗", 《电网技术》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112734028A (en) * | 2020-12-28 | 2021-04-30 | 三峡大学 | Modeling method for prediction model of concentration of dissolved gas in transformer oil |
| CN112749509A (en) * | 2020-12-30 | 2021-05-04 | 西华大学 | Intelligent substation fault diagnosis method based on LSTM neural network |
| CN113504423A (en) * | 2021-07-13 | 2021-10-15 | 许昌许继软件技术有限公司 | Primary equipment online monitoring data trend prediction method and device |
| CN113705086A (en) * | 2021-08-05 | 2021-11-26 | 陶帝文 | Ultra-short-term wind power prediction method based on Elman error correction |
| CN117667495A (en) * | 2023-12-29 | 2024-03-08 | 湖北华中电力科技开发有限责任公司 | Application system fault prediction method based on association rule and deep learning integrated model |
| CN117667495B (en) * | 2023-12-29 | 2024-07-05 | 湖北华中电力科技开发有限责任公司 | Association rule and deep learning integrated application system fault prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111060652A (en) | Method for predicting concentration of dissolved gas in transformer oil based on long-term and short-term memory network | |
| CN113889198B (en) | Transformer fault diagnosis method and equipment based on oil chromatography time-frequency domain information and residual error attention network | |
| CN104966161A (en) | Electric energy quality recording data calculating analysis method based on Gaussian mixture model | |
| CN112486137A (en) | Method and system for constructing fault feature library of active power distribution network and fault diagnosis method | |
| CN104281525A (en) | Defect data analytical method and method for shortening software testing programs by using same | |
| CN114418166A (en) | Method, device and medium for predicting concentration of dissolved gas in transformer oil | |
| Xing et al. | Health evaluation of power transformer using deep learning neural network | |
| CN115952404A (en) | Hydroelectric generating set fault early warning method, device and terminal | |
| CN117371207A (en) | Extra-high voltage converter valve state evaluation method, medium and system | |
| CN115859777A (en) | Method for predicting service life of product system in multiple fault modes | |
| CN116644358A (en) | Power system transient stability evaluation method based on Bayesian convolutional neural network | |
| CN113128396B (en) | Power quality composite disturbance classification method | |
| CN110516792A (en) | Non-stable time series forecasting method based on wavelet decomposition and shallow-layer neural network | |
| Ihsan et al. | Deep Learning Based Anomaly Detection on Natural Gas Pipeline Operational Data | |
| Fu et al. | PQEventCog: Classification of power quality disturbances based on optimized S-transform and CNNs with noisy labeled datasets | |
| Zhu et al. | Aiming to Complex Power Quality Disturbances: A Novel Decomposition and Detection Framework | |
| CN117743829A (en) | Short-term power load quantity prediction method based on deep learning | |
| CN116933162A (en) | Transformer running state determining method and device and electronic equipment | |
| CN112001530A (en) | Predictive maintenance method and system for transformer oil chromatography online monitoring device | |
| CN117454113A (en) | Power transformer fault early warning method based on CNN-BiLSTM and RF model | |
| CN116881798A (en) | Conditional gracile causal analysis method based on variable selection and reverse time lag feature selection for complex systems such as weather | |
| CN107085630A (en) | A kind of Gases Dissolved in Transformer Oil on-Line Monitor Device Analysis of The Practicability method | |
| Yeon et al. | Visual imputation analytics for missing time-series data in bayesian network | |
| CN116561569A (en) | Industrial power load identification method based on EO feature selection and AdaBoost algorithm | |
| CN112463584B (en) | Accurate test analysis method and device based on defect analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200424 |
|
| RJ01 | Rejection of invention patent application after publication |